Drones and Artificial Intelligence

While the introduction of programming revolutionized the application of human thought process to programs and machines, Artificial Intelligence might just be the means to close the gap between a programmed computer and the human brain. Artificial Intelligence or AI has enabled us to recreate, to an extent, the human thought process to programs that refine their decision making process and evolve with time to make better decisions. As the application of AI expands to a multitude of fields of applications; logistics systems, communication networks, manufacturing systems, security setups, traffic control etc. Its inclusion in every new and novel field of application is greatly contemplated and looked forward to.

One such novel field is drone logistics. Drones have been around for a lot of time, but due to the optimization of their control systems in the last decade, their domestic usage has risen up by a very steep curve. The article is focused on discussing the usage of AI on drone systems:

1.            How does it apply to drone systems?
2.            How does it increase a drone network’s efficiency?
3.            How does it expand endeavors possible with drone technology?

Intelligently Programmed Drones

Systems of drones operate in an organized way, with set time, paths for each drone, timely and trustworthy data transfer within the individual drones etc.

Most drones are operated by a pilot or user who controls the drone’s path and direction of the camera mounted on the drone. While it is applicable for photographers or people who use drones for recreational purposes, the commercialization of drone technology and its application in different markets has called for the need of automation for the following reasons:

1.    Repetitive tasks performed by drones
2.    Simplicity of tasks removing the need for a human operator
3.    Accuracy of performance of a programmed/fixed task.

Let us understand how autonomous drones work.

Having the program stored and communicated by a microcontroller, a single, individual drone is programmed to follow a set path according to a set of pre-programmed principles. This could be for example while detecting obstacles using sensors (ultrasonic, infrared etc.) to avoid obstacles, deliver its package or perform any other task it is being utilized for at the designated location. Whether planted in the drone or decided by the drone using the criteria defined by its programmer, the drone completes its mission and returns to its point of take-off or launch, all the while recording any data necessary.

Artificially Intelligent Drone Systems

Autonomous drone systems can work without AI, so why do we need it? Well, autonomous systems may not need a pilot to operate the drones all the time, but they do need a programmer. A programmer has to work all the time to make the drone’s operation more and more efficient and effective; artificial intelligence can move this responsibility to the drone itself and make it a lot more efficient as the drone will now observe the course of its operation and make changes in its program to achieve the same task allotted to it without the need for human involvement. Given enough tries or iterations, either in real time operation while performing the task or in virtual reality or simulations, the drone can improve upon its program again and again and move towards the best possible or the optimum solution or method for achieving any given task.

Let us take an example of a security drone that is designed to observe different people in a mall and classify them according to gender, age, height etc. The drone will be programmed with face detection and image recognition programs that are used to recognize and observe people with a camera. A person manually processing the images captured by the drone on the other hand, may have to work for countless hours to perfect this program to achieve accurate results. An artificially intelligent drone, however, will do it by itself; it will start observing facial hair and compare other features of specimens/people with facial hair with those without them and improve its classification on the basis of gender. It can ‘look’ at the skin of the people and their height to constantly improve its classification of people based on their age. This is an example of an artificially intelligent drone learning how to survey people.

A recent example of researchers using AI and drones to identify human behaviour used a combination of deep learning algorithms trained with 20,000 images and a support vector machine was recently published. The drones were able to successfully identify if two individuals were engaging in violent behaviour based on the angles of their limbs.

We can, therefore, identify some novel benefits of using artificial intelligence or machine learning technologies with drone systems:

1. Real time data: The data that would normally take days or weeks to be processed and made available can be filtered and processed by artificial intelligence in real time, continuously being updated and communicated.
2. Automated data capturing: Automated data processing opens up a new world for looking over complex physical structures and analyzing the captured data for defects. This can be applied to cell towers, wind turbines and general infrastructure inspection. Automated drones capture the images or other information required quickly, without having to deploy personnel to the top of the cell tower — today it is a manual process and requires individuals to access cell towers, thus exposing them to risks which potentially may cause fatalities. Automated drones can also scan blades on wind turbines without a human piloting the drone. Furthermore, it can carry trained AI models that can be used to quickly detect defects and provide recommendations on whether maintenance is necessary, without having to rely on archaic methods, such as a telescope.

2. Shortening construction project length: Drones combined with AI trained models for detecting all objects and activities inside a job site will eventually automate the process of understanding and predicting the daily changes on a construction site, and will give customers data to better plan and allocate resources more efficiently including optimizing routes for construction equipment travelling across the site and creating a daily deployment schedule for machinery and people.

3. Improving human operation: AI can measure the current efficiency of a driver or operator, and create a model of the most efficient operator from which newly hired operators can learn. It effectively transfers their experiences, through visual simulations, examples of hand movements, among other ways.

4. Public Safety: Through the use of object recognition and reiterative learning, drones may be used by law enforcement and public safety agencies to recognise and preempt risky, antisocial and violent behaviours.

Learning on the Job

There are a myriad of examples of drones using AI to improve efficiency and safety.

Skycatch, for example, have been operating AI drones on over 5,000 building sites in Japan, scanning the ground to create maps of the terrain, as well as the location of vehicles and and equipment. The drones are able to complete in 15 minutes what can take a team of humans several days.

The fixed wing UX11 drone developed by Delair uses deep learning to carry out in-flight calculations, speeding up the data processing chain when carrying out long-range mapping and surveying tasks. This improves workflow and reduces time spent covering large areas.

In Kyoto Japan, researchers used a publicly available deep learning package to process aerial images of photos taken by a UAV to identify tree crowns. With am 87% success rate, they showed that the tedious task of traipsing through forests to count individual trees could be sped up and made safer through use of drones and AI.

A Final Word

There are numerous application for drones and AI that will in time make daily tasks safer, more efficient and cheaper to perform. The question of ethics of some of these applications has been the subject of public interest. The recent involvement of Google’s Tensorflow in an AI development project with the US Department of Defense’s Project Maven drew considerable negative media attention as well as outcry from Google staffers. Google has since withdrawn from the project but it cannot be ignored that for as long as drones continue to be used for military purposes, they will draw upon AI to perform those missions with increasing accuracy.